CHINA’S DEEPSEEK RELEASES A 1.6 TRILLION PARAMETER AI MODEL THAT BENCHMARKS AGAINST GPT-5 AND GIVES IT TO THE WORLD FOR FREE DeepSeek released V4 in late April and the specifications have been reshaping assumptions inside the AI industry ever since. The flagship version, DeepSeek V4 Pro, contains 1.6 trillion parameters in a Mixture of Experts architecture that activates only 49 billion at any one time, keeping inference costs low while delivering benchmark results that sit alongside GPT-5.5 and Claude Opus 4.7 on agentic tasks. The model supports a one million token context window, among the largest available in any commercially usable system. The architecture uses a hybrid attention mechanism designed to improve efficiency on long documents, extended conversations, and complex multi-step reasoning chains. DeepSeek made V4 available for download and adaptation on Hugging Face under terms that allow commercial use. API pricing sits at $0.30 per million input tokens, a fraction of what American frontier labs charge for comparable capability. A smaller variant, V4 Flash, packs 284 billion parameters and targets speed and cost rather than raw performance. The strategic picture is hard to ignore. A Chinese AI lab operating under U.S. export controls on the most advanced chips is now producing models that benchmark against the best American systems and releasing them openly to the global developer community. The question of whether chip export restrictions meaningfully constrain China’s AI development capacity is one the results of this release increasingly answer in the negative. The gap between American and Chinese frontier AI capability, if it still exists, is closing. Keywords: DeepSeek V4, China AI model, open source AI 2026, DeepSeek benchmark